Automatic text location in images plays an important role in image content understanding

and draws attentions of researchers in the domain of computer vision. Current text location algorithms are mostly adaptive to specific applications; they are sensitive to the variation of text or images and lack robustness. This paper presents a universal approach for text location based on histogram features and AdaBoost. The new algorithm extracts histogram features

which have strong discriminabilities for text and non-text. AdaBoost algorithm with cascade structure is introduced to design the classifier for text texture. The algorithm transfers the binary output of the texture classifier into probability form and generates corresponding text probability image. CAMSHIFT algorithm is used to search for the final location result in the text probability image. The experimental results demonstrate the robustness of the proposed algorithm

which is adaptive to the text of different languages

fonts or scales

and gets promising location results in variant types of images.